-
Add changes
-
Update src/pipeline.py
-
Oops
-
Remove redundant
-
channels_last
-
q_ only
-
Update src/pipeline.py
-
Upload src/pipeline.py with huggingface_hub
-
Upload src/pipeline.py with huggingface_hub
-
Upload src/pipeline.py with huggingface_hub
-
Upload src/pipeline.py with huggingface_hub
-
Upload src/pipeline.py with huggingface_hub
-
Upload src/pipeline.py with huggingface_hub
-
Upload src/pipeline.py with huggingface_hub
-
Real base
-
Upload src/pipeline.py with huggingface_hub
-
Upload src/pipeline.py with huggingface_hub
-
Upload src/pipeline.py with huggingface_hub
-
Upload src/pipeline.py with huggingface_hub
-
Upload src/pipeline.py with huggingface_hub
-
Upload src/pipeline.py with huggingface_hub
-
Upload src/pipeline.py with huggingface_hub
-
Upload src/pipeline.py with huggingface_hub
-
Upload src/pipeline.py with huggingface_hub
-
Upload src/pipeline.py with huggingface_hub
-
Ok, compile
-
Tokenizer parallel
-
DiffusionPipeline
-
Baseline
-
COmpile is no longer the choice for this model
-
NO fullgraph
-
boring channels_first
-
quantize
-
Paste
-
Fix it
-
Channels_first
-
CUDA first
-
compile transformer
-
Fuse
-
Greedy
-
cudagraphs
-
Full
-
Channels_first
-
Run
-
eval
-
Good compile
-
Config without VAE
-
Advanced compile
-
compile vae
-
tiling